1. Field of the Invention
The present invention relates to a technique of manufacturing exposure masks used in photolithography or X-ray lithography and more specifically to a method of creating reduced projection printing mask data suitable for forming fine patterns by improving a way to correct design data.
2. Discussion of the Background
In recent years, the accuracy of pattern transfer in a lithographic process has been becoming a problem as the size of a device unit to be manufactured into large-scale integrated circuits becomes very small. More specifically, when a reduced pattern of a mask is transferred onto a wafer, there occur phenomena such that right-angled corners are rounded, line ends become short, and line widths increase/decrease, etc. Hereinafter, such phenomena are referred to as the optical proximity effect (OPE).
The causes of the OPE include optical factors (interference between light beams transmitted through adjacent patterns), resist processes (baking temperatures/times, development time, etc.), the reflection of light from and irregularities of the substrate, the effect of etching, etc. Though the causes include factors except the optical factors, the above phenomena are named as the optical proximity effect. When the absolute values of allowable dimensional errors decrease as patterns become finer, the allowable dimensional errors may be exceeded due to the OPE. The mainstream way of preventing pattern transfer degradation due to the OPE is to add corrections allowing for degradation to a mask in advance. Hereinafter, such processing is referred to as optical proximity correction (OPC).
So far, many OPC methods have been reported. These methods are roughly classified into rule-based approaches to make corrections on the basis of correction rules which have been obtained in advance and simulation-based approaches of using a simulator by which phenomena involved in exposure processes are modeled. Typical examples of these two approaches will be described below.
First examples of rule-based approaches are described in Oberdan W. Otto et al.: "Automated optical proximity correction--a rule-based approach", pp. 278-293, and Richard C. Henderson et al.: "Correcting for proximity effect widens process latitude", pp. 361-370, both in Optical/Laser Microlithography VII, Vol 2197, SPIE Symposium on Microlithography 1994. These approaches make corrections on a mask pattern in advance allowing for dimensional errors due to various factors as described above.
Corrected masks are such that a pattern is partly widened/narrowed (refer to FIG. 1), a corner stressing pattern is placed in corners (see FIG. 1), an assist feature or features of less than resolution limit are placed inside or outside a pattern (see FIG. 2). In FIG. 1, dotted lines indicate mask patterns prior to correction, while fine lines indicate mask patterns after correction. In FIG. 2, reference character P denotes light shading mask patterns, while AF denotes an assist feature of less than limit resolution. The assist features are known to have effects of improving the pattern transfer accuracy if they are combined with annular illumination and preventing the sidelobes of the main pattern from being resolved if they are combined with a halftone type of phase shift mask. It is also known that the resolution of the main pattern is improved by placing a phasing member relative to either of the assist feature and the main pattern so that the phase of light transmitted through the assist feature is shifted by approximately 180 degrees with respect to the phase of light transmitted through the main pattern.
The correction procedure in the conventional approaches includes creating correction rules in advance and correcting the mask pattern accordingly. For example, the amount of correction dE for the correction amount of a reference edge is represented by a combination of parameters L0, L1, L2, G0, G1, W0, W1, and W2 shown in FIG. 3 as follows: EQU dE=f(L0, G0, W0, L1, G1, W1, L2, W2)
dE values are calculated based on various combinations of the parameters and entered into a table. In correction, a search is made of the table for a match in parameter combination. With a combination that is not included in the table, dE is obtained by interpolating between members in the table.
As an example of a conventional simulation-based approach, an OPC method using simulation of an optical image is discussed in the paper entitled "Fast optical proximity correction: analytical method" by Satomi Shioiri et al., Optical/Laser Microlithography VIII, Vol. 2440, SPIE Symposium on Microlithography 1995, pp. 261-269. With this method, a rectangle of a width of .DELTA.t is added/removed along the contour of a mask pattern and the width .DELTA.t is sought so that the light intensity at a point of interest on the contour reaches a desired value. By using this method, it is expected that correction can be made in a time several times the time required to calculate the image intensity at a point of correction.
However, those conventional techniques have the following problems.
With the rule-based approach, in the event that the previously prepared rules cannot apply, it is required to obtain a correction value by means of interpolation. Thus, there arises the possibility of error. In addition, it is very troublesome to seek all the rules for any of layouts in a mask. With complicated two-dimensional layouts in particular, suitable parameters become difficult to set up and moreover the parameters increase in number.
With the simulation-based approach, on the other hand, the optical image simulation requires very lengthy calculations. At present, it is considered very difficult to simulate the entire area of an LSI chip that measures 1 to 2 centimeters on a side. In particular, logic devices in which mask patterns are not always hierarchically defined will have an enormous amount of area to be corrected as compared with memory devices which have hierarchically defined mask patterns. Thus, the problem of processing time is more serious.
Moreover, the simulation-based approach, which simply displaces the contours of pictorial characters, cannot create assist features of less than limit resolution inside or outside the pictorial characters.
Thus, of the conventional techniques of correcting the OPE in creating mask data, the rule-based approach has a problem that there arises the possibility of error when the correction rules do not apply and the simulation-based approach has a problem that an enormous amount of time is required.
Next, a description will be given of a conventional technique in which attention is paid to active gates in a device.
As to the gate layer in the logic section in an LSI device, the dimensional precision of the width of active gates has a great effect on the device performance such as operation speed. For this reason, a very great precision in dimension is required and the OPC has to be made with precision. The layout of each active gate is characterized in that it generally has a length that is sufficiently great relative to the width. In many cases, therefore, the OPC for active gates is made with respect to one dimension, i.e., in the direction of width, in which case the direction of length is not taken into consideration.
A third conventional technique which applies the OPC to the active gates in the logic section is described in an article entitled "Optical Proximity Correction, a First Look at Manufacturability" by Lars. W. Liebmann, Photomask Technology and Management, SPIE Vol. 2322, 1994, pp. 229-238. In this technique, the gates in a 64-Mbit DRAM are subjected to the OPC with respect to the direction of width.
The procedure of the OPE correction will be described here with reference to FIG. 4. In this figure, areas indicated by oblique lines extending up from right to left indicate a gate conductor layer, an area indicated by oblique lines extending up from left to right indicate a diffusion layer, dotted areas indicate active gates, and bold lines indicate sides to be corrected. Of the patterns of gate conductor layer, sides (edges) including the active gates are extracted. For the active gate edge 1, the spacing or distance (D) between the edge and the pictorial character that is closest to the edge is measured. The edge is then shifted by the amount of bias corresponding to the spacing. When the left edge 2 is made a candidate for correction, it is shifted by the amount of bias corresponding to D, the spacing between the edge 2 and the closest pattern. A relationship between the spacing and the amount of bias has been sought in advance in the form of a table as shown below and an edge is shifted for correction while the table is being referred to.
TABLE 1 ______________________________________ SPACING IN AMOUNT OF BIAS IN MICROMETERS (.mu.m) MICROMETERS (.mu.m) ______________________________________ -0.04 0.01 0.41-0.5 0.02 0.51- 0.03 ______________________________________
As a fourth example, the correction of gate width in logic devices is described in an article entitled "Simple Correcting Method of Optical Proximity Effect for 0.35 m Logic LSI's" by Eiich Kawamura et al, Proceedings of Microphotolithography Digest of Papers, 1995, pp. 286-287. Basically, this method is the same as the third technique described above in that the amount of bias corresponding to the distance to the adjacent pictorial character is added to an edge of interest.
The fourth method is distinct from the third method in that an edge is divided into line segments according to the position of a corner or corners of the adjacent pattern, and each line segment is shifted by the amount of bias corresponding to the distance to the adjacent pattern. In the example of FIG. 4, the edge 1 is divided into three segments S1, S2, and S3. For each of S1, S2 and S3, a search is made of the table for the amount of bias for the corresponding one of distances A, B and C. The edge segment is then shifted by the corresponding amount of bias.
In addition, the one-dimensional OPC approach is described in the article entitled "Automated Optical Proximity Correction--a Rule-Based Approach" by Oberdan W. Otto, which has been described herein as the first conventional technique.
As shown in FIG. 5, the amount of bias dE of an edge of interest is represented by a function of line/space parameters in the one-dimensional direction. In this case, the farther away the parameter is from the edge of interest, the less influence it will have on the amount of bias. In other words, the more the parameters, the more accurate dE will become. In the first conventional technique, a table is prepared in advance which establishes a correspondence between sets of parameters and amounts of correction. When no matching set of parameters is found in the table during correction processing, interpolation between sets of parameters in the table is performed to obtain the amount of correction.
The problems with the one-dimensional OPC in the first, third and fourth conventional techniques will be described with reference to FIGS. 6A, 6B, and 6C.
The OPC of layouts shown in FIGS. 6A, 6B and 6C using a commercially available program resulted in the optimum amounts of bias indicated in Table 2 below for edges (1), (2), (3) and (4). The exposure and mask conditions at this point were that annular illumination of a shading factor of 2/3 and a halftone mask of =0.6, NA=0.57, wavelength=365 mm, and amplitude transmittance factor=0.223607 were used.
TABLE 2 ______________________________________ AMOUNT OF BIAS Edge position FIG. 6A FIG. 6B FIG. 6C ______________________________________ (1) 0.012 0.012 0.012 (2) 0.012 0.009 0.009 (3) 0.012 0.005 0.005 (4) 0.012 0.028 0.028 ______________________________________
Let us consider the case where the mask shown in FIG. 6B is subjected to the OPC by means of the third and fourth conventional techniques. In the third and fourth techniques, the amount of bias is set according to the distance between a point of interest and its adjacent pattern. First, a rule is determined from the layout of FIG. 6A such that the amount of bias is 0.012 micrometers when the spacing between adjacent pattern is 0.525 micrometers, so that the amount of bias of 0.012 micrometers is applied to all of edges whose distance to their respective adjacent pattern is 0.525 micrometers.
As for the edge (1) shown in FIG. 6B, the optimum amount of bias is 0.012 micrometers. As for the edges (2), (3) and (4), however, it can be read from Table 2 that their optimum amount of bias is not 0.012 micrometers because they are affected by a large pattern that locates on the right side of them. That is, when lines and spaces are arranged regularly, it is sufficient to apply to each edge the amount of bias corresponding to the distance to its adjacent pattern. When lines and spaces vary in width, however, a problem arises in that the precision of correction is not sufficient.
In the one-dimensional OPC of the first conventional method, unlike the third and fourth methods, the number of parameters is increased. That is, by taking into consideration lines and spaces which are far from an edge of interest, the precision of correction can be increased, eliminating the problem with the third and fourth methods.
For this reason, in order to correct the edge (2) shown in FIG. 6B with a sufficient precision, dE, the amount of bias is set as a function of nine parameters, i.e., f(L0, G0, G1, L1, L2, G2, G3, L3, L4). The use of the nine parameters will result in the amount of bias of the same precision as in Table 2. As is understood from the fact that the amounts of bias for the layouts shown in FIGS. 6B and 6C are identical, any difference in the layout to the right of L2 has no effect on the amount of bias. It will therefore be understood that, for the edge (2), parameters associated with the layout to the right of the pattern L2 are unnecessary.
In general, contribution of another point to the amount of bias for a certain point of interest is determined by optical conditions, layout, and the distance between the two points. The contribution of a point that is far away from the point of interest can be ignored. By contrast, a point that is at a sufficiently short distance from the point of interest makes contribution to the amount of bias for the point of interest. Hereinafter, the area that is within such a distance from the point of interest is referred to as the OPE range.
In summary, the problems with the one-dimensional OPC in the first conventional technique are that, if the number of parameters is too small, the correction precision becomes insufficient for the same reason as in the fourth conventional technique and, on the other hand, if the number of parameters is increased too much so as to assure the precision, much time is required for rule calculations and a table that makes amounts of bias correspond with parameters becomes larger than is necessary.
Moreover, a problem arises in the first conventional technique in that a correction table must be prepared prior to correction processing. The preparation of a correction table for any layout involves a lot of work and the table also increases in size. Furthermore, when there is no matching parameter set in the correction table, interpolation is made between parameter sets in the table, so that an error may occur in the process of interpolation.
Next, conventional techniques related to two-dimensional OPC will be described.
Methods of making automatic OPC on design data are roughly classified into two types: the simulation-based approach, and the rule-based approach. The simulation-based approach is not suited to handle data in a large-scale layout because much time is required to simulate an optical image. Thus, the rule-based approach appears to be more practical.
At this point, the first conventional technique described previously will be described again. In the first conventional technique, the contour of an input design pattern is divided into line segments, and a point of correction is set at the midpoint of each line segment (refer to FIG. 7). In FIG. 7, a symbol .largecircle. indicates a point of correction. In division, each edge is divided at a point where a change occurs in the surrounding one-dimensional placement and the periphery of each corner is divided as well. For each point of correction, a selection is made among one-dimensional, 1.5-dimensional and two-dimensional rules according to its surroundings. The one-dimensional and 1.5-dimensional rules are applied to normal edge portions, while the two-dimensional rules are applied to corners. To make correction on each point of correction, reference is made to the correction table for the selected rules.
As a fifth conventional technique associated with the rule-based approach, there is a technique described in the article entitled "Large Area Optical Proximity Correction using Pattern Based Corrections" by David M. Newmark, Photomask technology and Management, Vol. 2322. In this technique, the correction is carried out by shifting with a proximity effect correction window (refer to FIG. 8) comprising a correction zone and a buffer zone. The buffer zone is placed around the correction zone to cover the proximity effect range. In shifting the proximity effect correction window, it is considered to be efficient to shift it so that its center will be placed at the corners and the centers of edges of a pattern (refer to FIG. 9). In FIG. 9, a symbol .largecircle. indicates the center of the proximity effect correction window.
A sixth conventional technique relating to rule-based correction of contact holes in particular is a method which corrects the ratio of longitudinal and lateral dimensions of a hole pattern in accordance with intervals at which the hole pattern is arranged, the method being disclosed in Japanese Unexamined Patent Publication No. 8-254812. In this technique, the ratio of longitudinal and lateral dimensions of a hole pattern is determined as a function of the pitch Px of the hole pattern placement in one (x) direction. This method cannot be applied to contact holes that are regularly placed two-dimensionally. The publication thus says that the pitch Py in the other (y) direction should preferably be larger than the OPE range, i.e., not less than 3.lambda./NA.
The problems with the conventional techniques will be described below.
In the first conventional technique, a reference is made to the table for each of points of correction set by dividing edges. In the case of FIG. 7, there are 20 points of correction and hence 20 references to the table must be made. In addition, since edges must be divided before processing, specific correction processing cannot be performed to suit the shape of a portion of a pattern or the shape of the entire pattern such as performing specific correction processing on a line end or performing specific correction processing on the shape of contact holes.
In the fifth conventional technique, correction is carried out while shifting an OPE window of a fixed size. In the case of FIG. 9, therefore, correction will be repeated four times for the pattern a and two times for the sides b-1 and b-2 of the pictorial character b. The sixth conventional technique cannot be applied to a two-dimensional placement of contact holes or to an aperiodic placement of contact holes.
Further, conventional techniques relating to two-dimensional OPC will be described next.
An example of a rule-based correction method is taught in the article entitled "Automated Optical Proximity Correction, a Rules-Based Approach", SPIE, vol. 2197, 1994, p 302. In this seventh conventional technique, a pattern placement is described by parameters (mainly pattern width, spacing, and length in one dimension). A table that describes correction values for this pattern placement is prepared and then referenced in performing correction processing. With this rule-based correction method, no straightforward method of describing and referencing a two-dimensional correction data table has been established. At present, simulation-based correction methods are mainly investigated.
An eighth conventional technique relating to the simulation-based correction is described in IEEE Trans. Electron Devices, Vol. 38, No. 12, 1991, p 2599, and uses an aerial image as a lithography model. To be specific, by performing an aerial image simulation on an input mask pattern, the amount of displacement of the optical image from a desired pattern edge is calculated and an edge of the mask pattern is shifted (corrected) by the amount of displacement in the direction opposite to the direction of displacement. The above operation is repeated to make the optical image approach the desired pattern.
As an example of making correction by taking all of process conditions into a model, there is the following report. That is, a test pattern is transferred onto a wafer under certain process conditions, a model (behavior model) is produced by measurements of pattern on the wafer, and thus all of process effects such as exposure, development, etching, etc., are included in the model. Correction values are calculated on the basis of that model (for example, SPIE vol. 2197, p 371). As to the simulation-based correction method, there are many other reports.
In applying the automated mask pattern correction technique to a large-area mask pattern, a method of making calculations and correction on all the area of the mask pattern at the same time is not practical because computers have limited processing speed and storage capacity. In order to use hardware resources efficiently, it will be easy to divide the mask pattern into areas suitable for correction and make correction on each of the areas on the basis of a model or correction rules.
Divided correction processing will be described with reference to FIG. 10. In this figure, P0 indicates a design layout area, P1 indicates a corrected layout area, A indicates an area to be corrected, A' indicates an area around the area to be corrected, a indicates an already corrected area, and Au indicates an uncorrected area. Assume that only the inside of one of divided areas, A, is set as an optical proximity effect calculation area for correction. Then, the OPE range from correction area in the surrounding divided areas A's will be disregarded. If, therefore, the area a thus corrected is incorporated into an already corrected mask pattern, the mask pattern will involve the results of improper correction. In particular, it is expected that the improper correction becomes noticeable in patterns in the neighborhood of the boundaries of A.
To avoid such improper correction, a method is described in SPIE Vol. 2197, p 348, which adds a buffer area to the periphery of a to-be-corrected area. This method is illustrated in FIG. 11. A buffer area B which covers the OPE range from the surrounding areas is added to the periphery of a to-be-corrected area A cut out from the design pattern P0 and the to-be-corrected area A and the buffer area B are combined to form an optical proximity effect calculation area C. Calculations are performed on the area C to obtain correction values. Among correction solutions obtained by the calculations, a correction solution for the to-be-corrected area (a) is returned to the corrected mask pattern P1. The procedure then goes to calculations on the next area. In FIG. 11, B is the buffer area of A, C is the OPE calculation area, c is an already corrected optical proximity effect calculation area, b is the buffer area of c, and a is the corrected solution of the to-be-corrected area.
FIGS. 12A and 12B are enlarged views of the area A and its surrounding area of FIG. 11. The to-be-corrected area A is surrounded by already corrected areas (1) to (4) and uncorrected areas (5) to (8). More specifically, FIG. 12A shows a to-be-corrected area and a buffer area in the conventional correction method, while FIG. 12B shows a to-be-corrected area and a buffer area which are actually used. As shown in FIG. 12A, a pattern input to a correction system as an optical proximity effect calculation area is an uncorrected design pattern containing a to-be-corrected area A and a buffer area B regardless of the progress of correction of the entire area of the mask pattern. For this reason, the buffer area of each to-be-corrected area should have contained already corrected patterns b and uncorrected patterns B from the point of view of the progress of correction. However, since all of uncorrected patterns B are input as a buffer area in performing calculations on each OPE calculation area, an accurate correction solution cannot be obtained within the to-be-corrected area when the pattern within the buffer area differs greatly before and after correction.
In FIG. 13, a mask pattern to be corrected is shown divided into areas A1 to A4. More particularly, this figure illustrates how many times correction calculation is performed repeatedly on each area. Some part in the area is calculated repeatedly because a buffer area of each divided area overlaps another divided area. Lb indicates the range of a buffer area, and S1 to S4 indicate areas in which correction calculations are performed one time to four times. As shown in FIG. 13, since the correction calculations are performed on the entire OPE calculation area for each to-be-corrected area, uncorrected patterns in the buffer areas are subjected to correction calculation and pattern data processing two (S2) to four times (S4), resulting in waste of a large amount of calculation time.
In dividing a mask pattern of a large area and making correction for each divided area, the conventional method inputs uncorrected pattern as a buffer area for correction.
The first problem with this method is that, even for an area for which correction calculations have already performed and a correction solution has been obtained, an uncorrected pattern is taken as a buffer area into an OPE calculation area. And hence the pattern taken into the buffer area differs from a pattern to be placed after the termination of correction operation. For this reason, a deviation from a true correction solution will occur. The second problem is that, when dividing correction is made on a mask pattern of a large area, calculations are made redundantly for buffer areas, resulting in waste of calculations.